局部鉴别分析驱动的红外与可见光图像协同目标跟踪

Infrared and Visible Fusion for Robust Object Tracking via Local Discrimination Analysis

针对红外图像所含信息量少、可见光图像易受环境影响的问题,提出一种基于局部鉴别分析的红外与可见光图像多源信息协同跟踪的目标跟踪方法.从评估图像信息对目标和背景间的可区分性能角度出发引入线性鉴别分析理论,建立了局部区域目标背景间的可区分度函数;以此为基础实现了多源图像在特征层次上的自适应融合;最后将该融合理论嵌入到粒子滤波跟踪框架中,实现对目标的跟踪.实验结果表明,与采用单一图像信息的目标跟踪系统相比,该方法可对红外和可见光图像进行有效融合,实现对目标的稳健跟踪.

The infrared images generally contain less information but the visible images are easily affected by environments.To address this problem,a local discrimination analysis based infrared and visible multi-source information cooperative tracking approach is presented in this paper.From the view of evaluating the image information＇s ability of distinguishing the object form background,the Fisher linear discrimination theory is introduced to design the discriminative function between the target and background in local regions.Based on this,the fusion of multi-source image is executed adaptively on the feature level.Finally,we incorporate the proposed fusion method into the particle filter tracking framework to achieve the object tracking.Experimental results demonstrates that,compared with the tracking system with single image source,the proposed algorithm can effectively fuse the infrared and visible images to reliably track the object.